SAIS: Supervising and Augmenting Intermediate Steps for Document-Level Relation Extraction

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2395 - 2409 July 10-15, 2022 ©2022 Association for Computational Linguistics SAIS: Supervising and Augmenting Intermediate Steps for Document-Level Relation Extraction Yuxin Xiao1, Zecheng Zhang2, Yuning Mao3, Carl Yang4, Jiawei Han3 1Carnegie Mellon University, 2Stanford University, 3University of Illinois at Urbana-Champaign, 4Emory University 1yuxinxia@cs.cmu.edu 2zecheng@cs.stanford.edu 3{yuningm2,hanj}@illinois.edu 4j.carlyang@emory.edu Abstract Stepping from sentence-level to document- level, the research on relation extraction (RE) confronts increasing text length and more com- plicated entity interactions. Consequently, it is more challenging to encode the key infor- mation sources—relevant contexts and entity types. However, existing methods only implic- itly learn to model these critical information sources while being trained for RE. As a result, they suffer the problems of ineffective super- vision and uninterpretable model predictions. In contrast, we propose to explicitly teach the model to capture relevant contexts and entity types by Supervising and Augmenting Intermediate Steps (SAIS) for RE. Based on a broad spectrum of carefully designed tasks, our proposed SAIS method not only extracts relations of better quality due to more effec- tive supervision, but also retrieves the corre- sponding supporting evidence more accurately so as to enhance interpretability. By assess- ing model uncertainty, SAIS further boosts the performance via evidence-based data augmen- tation and ensemble inference while reduc- ing the computational cost. Eventually, SAIS delivers state-of-the-art RE results on three benchmarks (DocRED, CDR, and GDA) and outperforms the runner-up by 5.04% relatively in F1 score in evidence retrieval on DocRED.1 1 Introduction Playing a crucial role in the continuing effort of transforming unstructured text into structured knowledge, RE (Bach and Badaskar, 2007) seeks to identify relations between an entity pair based on a given piece of text. Earlier studies mostly pay attention to sentence-level RE (Zhang et al., 2017; Hendrickx et al., 2019) (i.e., the targeting entity pair co-occur within a sentence) and achieve promising results (Zhang et al., 2019; Zhou et al., 2020). Based on an extensive empirical analysis, 1Our code is available at https://github.com/ xiaoyuxin1002/SAIS. Peng et al. (2020) reveals that textual contexts and entity types are the major information sources that lead to the success of prior approaches. Given that more complicated relations are often expressed by multiple sentences, recent focus of RE has been largely shifted to the document level (Yao et al., 2019; Cheng et al., 2021). Existing document-level RE methods (Zeng et al., 2020; Zhou et al., 2021) utilize advanced neural architec- tures such as heterogeneous graph neural networks (Yang et al., 2020) and pre-trained language models (Xu et al., 2021b). However, although documents typically include longer contexts and more intri- cate entity interactions, most prior methods only implicitly learn to encode contexts and entity types while being trained for RE. As a result, they deliver inferior and uninterpretable results. On the other hand, it has been a trend that many recent datasets support the training of more power- ful language models by providing multi-task anno- tations …

Content truncated to 500 words